VMware Cloud Community
Dell_SRM_User
Contributor
Contributor

ESXi 5.1 Distributed Switch "creating a mac for every vlan on every nic for every host"

Good morning all,

the subject line statement was given to me in an email from our network authority... is it true???

I have basically come in to a company (as a contractor) to assist with the existing vSphere 5.1 environment. An upgrade to at least version 5.5 will be on the cards once the hardware has been upgraded..

However in the meantime when I have broached the question on to why some ESXi hosts do not have uplinks connected to the vDS that was configured by another contractor before me. The answer above in the subject line is what I got back, as it seemes there was an outage if all ESXi hosts had their uplinks connected to the vDS.

I have never come across this type of behaviour before with a vDS, I cannot find anyone at all on the VMware knowledge base with a similar issue so i wanted to ask the community is this a correct diagnosis? (I'll post more of the conversant below for context).

Any help or suggestions greatly appreciated..    email conversation below

"xxxxxx created distributed switches across the clusters, what this has done is create a mac for every vlan on every nic for every host, that’s currently 5500 mac with the two nics allocated, that’s before even a single vm has been allocated.

The existing vswitch seems to create a mac in the vlan only if a vm is created in the vlan (switch + vm mac) and also seems to have an issue with removing mac’s when moved/deleted (host reboot clears)

So basically we are in trouble with switch capacity, if we continue with the plan of 4 nics per host that will be 11000 mac addresses before a single vm is allocated (or french vlans added), clearly a problem. (approx. 5000 per access switch too)

The next switch model up (3850) can support 32k mac addresses, which would be a minimum I would consider acceptable, the current access switches in theory could be switched into  an l2 mode allowing 12k mac’s which *may* also be able to cover us"


Tags (4)
0 Kudos
1 Reply
Gortee
Hot Shot
Hot Shot

I will need additional information to be really helpful.  But the first place to look is in the switch load balancing algoritm I am willing to bet they have it set for IP-Hash take a look at this older article on the methods:

http://blog.jgriffiths.org/?p=866

If it is set for IP-Hash and they are not using port channels on the switch this could potentially cause this issue if both links are in the mix. 

The information that would be helpful is the following:

-Load Balancing policy on the virtual switch

-Number of nic's on ESXi host

-Number of VM's total

-Number of switches that nic's are connected to

Provide this information and I'll see what I can do to help

Thanks,

J

Joseph Griffiths http://blog.jgriffiths.org @Gortees VCDX-DCV #143